Ankit Kapoor

I investigate
how AI
changes
competition,
customer value,
and
business
economics.

Ankit Kapoor. Former machine learning engineer. MBA candidate at USC Marshall.

I build experiments and decision tools to make strategic assumptions easier to question. This portfolio follows three questions: how businesses respond to AI, where their advantage survives, and what the financial results need to be.

View the workAbout me

Selected work

Three independent projects. Each states its status plainly: all three run as software you can open and use, and only one has recorded a measured result.

Project 01

Disrupt This Business

How does AI change competition?

Concept preview

Concept preview — the decision scene the game presents: two positions, two customer segments, four quarterly commitments. Parameters are fictional scenario values from the project brief; no result is shown.

Drawn wider than this screen — scroll the figure sideways, or read it in words below.

What this figure shows, in words
  • Position A sells seats: priced per team per year, with support cost rising for each team added. Its stated advantage is the installed base.
  • Position B sells completed work: priced per delivered bundle, with cost falling as delivery automates. Its stated advantage is unit cost.
  • Both compete for the same fictional market of 800 small teams, which are price sensitive and cheap to switch, and 200 enterprise teams, which weigh reliability first and are costly to switch.
  • Play runs over four quarters. Each quarter the player commits one move, the opponent's move is then revealed, and the result resolves from stated rules.
  • After the result, the player switches sides and attacks the strategy they just recorded.
  • No outcome, score, or currency figure is shown: this is a drawing of the decision the game presents, not a record of a result it produced.

Competitive strategy scenario

A competitive strategy game about defending an established business or building the challenger. Explore decisions, inspect their consequences, and switch sides. The scenario is a fictional project-management market: an incumbent selling seats against a challenger selling completed work. Four quarterly commitments resolve through a deterministic economic model, so every outcome can be traced back to the rule that produced it.

Status: PrototypeEvidence: Illustrative

Read the case: Disrupt This BusinessPlay the scenario (opens in a new tab)

Project 02

The Moat Test

What remains worth paying for?

Concept preview

Concept preview — the comparison the investigation is built around: one transcript, two extraction methods, and a line reference behind every claim. The excerpt text is hand-authored placeholder content, not a recorded run.

Drawn wider than this screen — scroll the figure sideways, or read it in words below.

What this figure shows, in words
  • The source is a transcript with speaker labels and stable line IDs, shown here as eight lines. Two of them are explicitly tagged: line 3 reads “DECISION: ship Thursday” and line 6 reads “ACTION: Priya to draft”.
  • The baseline method extracts only tagged lines. It returns “ship Thursday” from line 3 and “Priya to draft” from line 6, and ignores every untagged line. It is a deterministic string match and sets the floor any model has to beat.
  • The challenger method returns richer structure: a decision (ship Thursday, current, from line 3), an action (Priya, due unresolved, from line 6), an open question (who signs off, from lines 4 to 5), and one item where the owner was ambiguous and was therefore left null, from line 7.
  • Every claim in either panel carries the transcript line it came from. Both panels stay blind until the reveal.
  • No accuracy figure appears in the drawing. The figures from the recorded baseline runs are given as text under Evidence and results, where their sample size and caveats can travel with them.

Product investigation with a working challenger

An investigation into what makes an AI product worth paying for. Build a small alternative, test its limits, and examine the business advantages the prototype leaves unresolved. The first case takes meeting assistants: if a model can summarise a transcript, what is the subscription actually buying? Every output is traced to the transcript lines behind it, and each claim carries its data mode.

Status: PrototypeEvidence: Measured, partial

Read the case: The Moat TestTry the challenger (opens in a new tab)

Project 03

Priced In

What do the economics require?

Concept preview

Concept preview — the expectations map the workbench produces: growth and margin combinations consistent with one valuation target. Axes are unitless and the target is hypothetical; no company, price, or forecast is shown.

Drawn wider than this screen — scroll the figure sideways, or read it in words below.

What this figure shows, in words
  • The horizontal axis is revenue growth and the vertical axis is operating margin. Both are unitless and carry no tick values.
  • A band crosses the grid from upper left to lower right. Every point on it satisfies the same hypothetical enterprise-value target, so higher required growth trades against lower required margin.
  • Two points on the band are annotated. One sits at modest growth with high margin; the other at fast growth with thin margin. They are alternative futures consistent with the same target, not a prediction of either.
  • Selecting a point feeds the operating bridge, which states three requirements: the customers the business must add, the price it must hold per customer, and the contribution the AI initiative must make.
  • There is no company, price, forecast, or return in the drawing, and the target is hypothetical.

Reverse-valuation workbench

A financial workbench for exploring the business performance required to justify a valuation. Connect assumptions to operating requirements and test the contribution an AI investment would need to make. It runs the valuation backwards: pick a target, see which growth and margin combinations satisfy it, translate one of those into required customers and price, and then try to break the case you just built.

Status: PrototypeEvidence: Illustrative

Read the case: Priced InExplore the sample (opens in a new tab)

How I approach a question

The same three moves, in the same order, on every project.

  1. Step 01

    Investigate the mechanism

    Work out how the money actually moves before deciding whether the technology matters, because most AI arguments are really arguments about a cost structure nobody has written down.

  2. Step 02

    Test the assumptions

    Build the smallest thing that could show an assumption is wrong, then run it against a baseline dull enough that beating it means something.

  3. Step 03

    Make the trade-off explicit

    State what the recommendation costs and what would change it, so the reader is disagreeing with a position rather than with a tone.

About

Read the full biography

My background is in machine learning engineering, and I am now pursuing an MBA at USC Marshall. I am interested in the decisions around AI: where it creates value, what makes that value defensible, and how to judge whether an investment is worthwhile.

These projects are a way to investigate those questions in public, with assumptions and limitations open to inspection.

  • USC MarshallMBA candidate
  • BITS Pilani DubaiComputer science

Contact

Contact details will be added before launch.